Failure due to a shape mismatch by inference.py

[yukisoya]

I executed the following command in a container created from the provided Dockerfile:

micromamba run -n diffdock python -m inference --config default_inference_args.yaml --protein_ligand_csv data/testset_csv_2.csv --out_dir results/user_predictions_small

The testset_csv_2.csv used for inference contains the following data:

complex_name,protein_path,ligand_description,protein_sequence
0,data/PDBBind_processed/6qqw/6qqw_protein_processed.pdb,data/PDBBind_processed/6qqw/6qqw_ligand.mol2,

In addition, I've made a modification to the inference_utils.py script to ensure proper handling of the complex_name column in the CSV file. Initially, I encountered an error when the complex_name values were integers. To resolve this, I edited the line 159 in inference_utils.py to labels.extend([str(complex_names[i]) + '_chain_' + str(j) for j in range(len(s))]).

However, I encountered a tensor shape mismatch error during the computation, as detailed in the output below:

DiffDock will run on cuda
Generating ESM language model embeddings
/home/appuser/micromamba/envs/diffdock/lib/python3.9/site-packages/torch/cuda/init.py:155: UserWarning:
NVIDIA H100 80GB HBM3 with CUDA capability sm_90 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70 sm_75 sm_80 sm_86.
If you want to use the NVIDIA H100 80GB HBM3 GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/

Processing 1 of 1 batches (2 sequences)
HAPPENING | confidence model uses different type of graphs than the score model. Loading (or creating if not existing) the data for the confidence model now.
/home/appuser/micromamba/envs/diffdock/lib/python3.9/site-packages/torch/jit/_check.py:181: UserWarning: The TorchScript type system doesn't support instance-level annotations on empty non-base types in __init__. Instead, either 1) use a type annotation in the class body, or 2) wrap the type in torch.jit.Attribute.

Size of test dataset: 1
0it [00:00, ?it/s]@> 6214 atoms and 1 coordinate set(s) were parsed in 0.03s.
/home/appuser/DiffDock/datasets/parse_chi.py:91: RuntimeWarning: invalid value encountered in cast
Y = indices.astype(int)
@> 6214 atoms and 1 coordinate set(s) were parsed in 0.03s.
Failed on tensor([0]) shape mismatch: value tensor of shape [196] cannot be broadcast to indexing result of shape [88]
1it [00:00, 1.62it/s]
Failed for 1 complexes
Skipped 0 complexes
Results are in results/user_predictions_small

I'm seeking help to resolve the tensor shape mismatch issue. Any advice on how to adjust the input data or modify the inference settings to prevent this error would be greatly appreciated.

Thank you for your support!

[gcorso]

Hi @yukisoya could you check whether the issue is still there with the fixes of the recent commits?